CA3¶
Imports¶
In [2]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier, IsolationForest
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import GridSearchCV
Reading data¶
In [17]:
df = pd.read_csv("./assets/train.csv", index_col = 0)
df.info()
# Checking for missing values
print(f"\nMissing values inn training data: {df.isnull().sum().sum()}")
<class 'pandas.core.frame.DataFrame'> Index: 2800 entries, -1.8257343 to -1.6260979 Data columns (total 9 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 Weight 2800 non-null float64 1 Sweetness 2800 non-null float64 2 Softness 2800 non-null float64 3 HarvestTime 2800 non-null float64 4 Ripeness 2800 non-null float64 5 Acidity 2800 non-null float64 6 Peel Thickness 2800 non-null float64 7 Banana Density 2800 non-null float64 8 Quality 2800 non-null int64 dtypes: float64(8), int64(1) memory usage: 218.8 KB Missing values inn training data: 0
Data exploration and visualisation¶
In [4]:
print(df.describe()) # Gives a table of the dataset with statistical components like mean and quantiles
sns.pairplot(df) # Gives a scatter plot for every pair & a histogram for the the feature paired with itself
Weight Sweetness Softness HarvestTime Ripeness \
count 2800.000000 2800.000000 2800.000000 2800.000000 2800.000000
mean -0.751050 -0.751005 -0.019557 -0.700683 0.771011
std 2.006590 1.955109 2.076865 2.029916 2.098275
min -7.103426 -6.434022 -6.959320 -7.570008 -7.423155
25% -2.238843 -2.104742 -1.593816 -2.112747 -0.572589
50% -0.882387 -0.997902 0.220174 -0.856858 0.930927
75% 0.853566 0.334989 1.542899 0.628895 2.229410
max 5.679692 6.438196 8.241555 5.942060 7.077372
Acidity Peel Thickness Banana Density Quality
count 2800.000000 2800.000000 2800.000000 2800.000000
mean -0.000989 0.506758 49397.491271 0.506429
std 2.286725 0.291936 29327.077623 0.500048
min -8.226977 0.000086 -980.343999 0.000000
25% -1.608385 0.257860 24025.427350 0.000000
50% 0.073963 0.506282 49303.534616 1.000000
75% 1.662417 0.761016 75066.598785 1.000000
max 6.395850 0.999430 99982.761410 1.000000
Out[4]:
<seaborn.axisgrid.PairGrid at 0x2d288a22610>
The pairs of different features does not look linear seperable, they have a complex relation.
Therefore, we think models like logistic regression will have low accuracy for theese data.
In [5]:
""" Checking correlation with every feature """
fig, axes = plt.subplots(1, 1, figsize=(14, 6)) # To make the values more readable
# annot = true to get value, coolwarm cmap to make sense with correlation
sns.heatmap(df.corr(), annot = True, cmap = "coolwarm")
Out[5]:
<Axes: >
As we can see: Acidity and Weight, and Weight and Sweetness has the highest positive correlation.
While the Acidity and Ripeness, and Softness and Ripeness has the highest negative correlation.
To conclude:¶
The pairs of different features have complex relations to each other. The data is not linearly seperable and
therefore, we think models like logistic regression will have low accuracy for theese data.
Acidity and Weight, Weight and Sweetness has the highest positive correlation.
While the Acidity and Ripeness, and Softness and Ripeness has the highest negative correlation.
With this informasjotion we know which features we can focus more on.
We also found out that the banana density values are too high. We want to scale and standardize
the values of the dataset to fit the models better.
We will probably remove the coloumns banana density and peel thickness because of the low correlation
to the other features.
Data Preprocessing & Feature Engineering¶
We found out in the exploration and visualization part that there were not any missing values.
However, we will still drop outliers if we find any.
Feature Selection: We saw that Banana Density and Peel Thickness has low correlation with other features.
Therefore we will drop those features
In [19]:
# Feature Selection, we are dropping the features when we're dropping coloumns in splitting data
useless_features = ["Banana Density", "Peel Thickness"]
# Including outliers, we found out that our model gets worse by removing outliers
df_clean = df.copy()
X = df_clean.drop(columns = ["Quality"] + useless_features)
y = df_clean["Quality"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.25, random_state = 42)
Data preprocessing and visualisation¶
In [20]:
# Standardizing
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test) # Using transform to prevent data leakage
print("Means:", X_train_scaled.mean(axis = 0))
print("\nStandard deviations:", X_train_scaled.std(axis = 0))
# Final Visualization
df_train_scaled = pd.DataFrame(X_train_scaled, columns = df_clean.drop(columns = ["Quality"] + useless_features).columns)
# Violin plots to visualize distribution
plt.figure(figsize = (12, 6))
sns.violinplot(data = df_train_scaled)
plt.xticks(rotation=60)
plt.show()
# Heatmap for visualizing correlation after data cleaning.
sns.heatmap(df_train_scaled.corr(), annot = True, cmap = "coolwarm")
# Pairplot for visualizing relations again.
sns.pairplot(df_train_scaled)
Means: [-1.94553368e-17 1.77635684e-17 -2.62224105e-17 1.69176842e-17 -2.07241631e-17 0.00000000e+00] Standard deviations: [1. 1. 1. 1. 1. 1.]
Out[20]:
<seaborn.axisgrid.PairGrid at 0x2d2a5bc7610>
It looks pretty much the same. However the correlation has just changed a bit.
For example weight-adicity correlation changed with -0.01.
Modelling¶
Finding good parameters¶
In [9]:
# Logistic Regression Classifier
param_grid_LR = {
'C': [0.01, 0.1, 1, 10, 100],
}
grid_search_LR = GridSearchCV(LogisticRegression(), param_grid_LR, cv=5, n_jobs=-1)
grid_search_LR.fit(X_train_scaled, y_train)
print(grid_search_LR.best_params_)
{'C': 1}
In [10]:
# SVC
param_grid_SVC = {
'C': [0.01, 0.1, 1, 10, 11],
'kernel': ['linear', 'rbf', 'poly'],
'gamma': ['scale', 'auto'],
}
grid_search_SVC = GridSearchCV(SVC(), param_grid_SVC, cv=5, n_jobs=-1)
grid_search_SVC.fit(X_train_scaled, y_train)
print(grid_search_SVC.best_params_)
{'C': 11, 'gamma': 'scale', 'kernel': 'rbf'}
In [11]:
# DecisionTree
param_grid_DT = {
'max_depth': [3, 10, 20, 30],
'min_samples_split': [2, 5, 7, 8, 9, 10],
'criterion': ['gini', 'entropy']
}
grid_search_DT = GridSearchCV(DecisionTreeClassifier(), param_grid_DT, cv=5, n_jobs=-1)
grid_search_DT.fit(X_train_scaled, y_train)
print(grid_search_DT.best_params_)
{'criterion': 'entropy', 'max_depth': 10, 'min_samples_split': 5}
In [12]:
# Random Forest
param_grid_RF = {
'n_estimators': [100, 125, 200],
'max_depth': [15, 20, 21, 25],
'criterion': ['gini', 'entropy']
}
grid_search_RF = GridSearchCV(RandomForestClassifier(), param_grid_RF, cv=5, n_jobs=-1)
grid_search_RF.fit(X_train_scaled, y_train)
print(grid_search_RF.best_params_)
{'criterion': 'gini', 'max_depth': 15, 'n_estimators': 200}
In [13]:
# KNN
param_grid_KNN = {
'n_neighbors': [3, 5, 7, 10, 15],
'p': [1, 2, 3, 4, 5, 6, 7]
}
grid_search_KNN = GridSearchCV(KNeighborsClassifier(), param_grid_KNN, cv=5, n_jobs=-1)
grid_search_KNN.fit(X_train_scaled, y_train)
print(grid_search_KNN.best_params_)
{'n_neighbors': 5, 'p': 5}
Finding the best model¶
In [14]:
dataset_sizes = np.arange(10, 2800, 50)
X_train_scaled = np.array(X_train_scaled)
y_train = np.array(y_train)
models = []
# Playing around with the variables to find something good
classifiers = {
"LogisticRegression": LogisticRegression(C = 1),
"SVC": SVC(C = 11, gamma = "scale", kernel = "rbf"),
"DecisionTree": DecisionTreeClassifier(criterion = "entropy", max_depth = 10, min_samples_split = 2),
"RandomForest": RandomForestClassifier(criterion = "gini", max_depth = 20, n_estimators = 125),
"KNN": KNeighborsClassifier(n_neighbors = 5, p = 5),
}
keys = classifiers.keys()
for clf_name in keys:
clf = classifiers[clf_name]
clf_index = list(keys).index(clf_name)
for size_index, size in enumerate(dataset_sizes):
X_train_subset, y_train_subset = X_train_scaled[:size], y_train[:size]
clf.fit(X_train_subset, y_train_subset)
y_pred = clf.predict(X_test_scaled)
accuracy = accuracy_score(y_test, y_pred)
models.append((clf, accuracy, size))
print(f"{clf_name} | Size: {size} | Accuracy: {accuracy:.4f}")
LogisticRegression | Size: 10 | Accuracy: 0.7857 LogisticRegression | Size: 60 | Accuracy: 0.8600 LogisticRegression | Size: 110 | Accuracy: 0.8671 LogisticRegression | Size: 160 | Accuracy: 0.8643 LogisticRegression | Size: 210 | Accuracy: 0.8614 LogisticRegression | Size: 260 | Accuracy: 0.8600 LogisticRegression | Size: 310 | Accuracy: 0.8643 LogisticRegression | Size: 360 | Accuracy: 0.8657 LogisticRegression | Size: 410 | Accuracy: 0.8600 LogisticRegression | Size: 460 | Accuracy: 0.8600 LogisticRegression | Size: 510 | Accuracy: 0.8600 LogisticRegression | Size: 560 | Accuracy: 0.8586 LogisticRegression | Size: 610 | Accuracy: 0.8614 LogisticRegression | Size: 660 | Accuracy: 0.8657 LogisticRegression | Size: 710 | Accuracy: 0.8629 LogisticRegression | Size: 760 | Accuracy: 0.8657 LogisticRegression | Size: 810 | Accuracy: 0.8657 LogisticRegression | Size: 860 | Accuracy: 0.8700 LogisticRegression | Size: 910 | Accuracy: 0.8700 LogisticRegression | Size: 960 | Accuracy: 0.8671 LogisticRegression | Size: 1010 | Accuracy: 0.8629 LogisticRegression | Size: 1060 | Accuracy: 0.8614 LogisticRegression | Size: 1110 | Accuracy: 0.8629 LogisticRegression | Size: 1160 | Accuracy: 0.8657 LogisticRegression | Size: 1210 | Accuracy: 0.8657 LogisticRegression | Size: 1260 | Accuracy: 0.8671 LogisticRegression | Size: 1310 | Accuracy: 0.8686 LogisticRegression | Size: 1360 | Accuracy: 0.8671 LogisticRegression | Size: 1410 | Accuracy: 0.8671 LogisticRegression | Size: 1460 | Accuracy: 0.8686 LogisticRegression | Size: 1510 | Accuracy: 0.8686 LogisticRegression | Size: 1560 | Accuracy: 0.8671 LogisticRegression | Size: 1610 | Accuracy: 0.8686 LogisticRegression | Size: 1660 | Accuracy: 0.8700 LogisticRegression | Size: 1710 | Accuracy: 0.8700 LogisticRegression | Size: 1760 | Accuracy: 0.8700 LogisticRegression | Size: 1810 | Accuracy: 0.8729 LogisticRegression | Size: 1860 | Accuracy: 0.8729 LogisticRegression | Size: 1910 | Accuracy: 0.8700 LogisticRegression | Size: 1960 | Accuracy: 0.8729 LogisticRegression | Size: 2010 | Accuracy: 0.8714 LogisticRegression | Size: 2060 | Accuracy: 0.8729 LogisticRegression | Size: 2110 | Accuracy: 0.8714 LogisticRegression | Size: 2160 | Accuracy: 0.8714 LogisticRegression | Size: 2210 | Accuracy: 0.8714 LogisticRegression | Size: 2260 | Accuracy: 0.8714 LogisticRegression | Size: 2310 | Accuracy: 0.8714 LogisticRegression | Size: 2360 | Accuracy: 0.8714 LogisticRegression | Size: 2410 | Accuracy: 0.8714 LogisticRegression | Size: 2460 | Accuracy: 0.8714 LogisticRegression | Size: 2510 | Accuracy: 0.8714 LogisticRegression | Size: 2560 | Accuracy: 0.8714 LogisticRegression | Size: 2610 | Accuracy: 0.8714 LogisticRegression | Size: 2660 | Accuracy: 0.8714 LogisticRegression | Size: 2710 | Accuracy: 0.8714 LogisticRegression | Size: 2760 | Accuracy: 0.8714 SVC | Size: 10 | Accuracy: 0.9000 SVC | Size: 60 | Accuracy: 0.9171 SVC | Size: 110 | Accuracy: 0.9200 SVC | Size: 160 | Accuracy: 0.9357 SVC | Size: 210 | Accuracy: 0.9529 SVC | Size: 260 | Accuracy: 0.9543 SVC | Size: 310 | Accuracy: 0.9500 SVC | Size: 360 | Accuracy: 0.9514 SVC | Size: 410 | Accuracy: 0.9571 SVC | Size: 460 | Accuracy: 0.9586 SVC | Size: 510 | Accuracy: 0.9657 SVC | Size: 560 | Accuracy: 0.9629 SVC | Size: 610 | Accuracy: 0.9671 SVC | Size: 660 | Accuracy: 0.9657 SVC | Size: 710 | Accuracy: 0.9643 SVC | Size: 760 | Accuracy: 0.9686 SVC | Size: 810 | Accuracy: 0.9700 SVC | Size: 860 | Accuracy: 0.9700 SVC | Size: 910 | Accuracy: 0.9671 SVC | Size: 960 | Accuracy: 0.9657 SVC | Size: 1010 | Accuracy: 0.9643 SVC | Size: 1060 | Accuracy: 0.9671 SVC | Size: 1110 | Accuracy: 0.9714 SVC | Size: 1160 | Accuracy: 0.9686 SVC | Size: 1210 | Accuracy: 0.9714 SVC | Size: 1260 | Accuracy: 0.9714 SVC | Size: 1310 | Accuracy: 0.9714 SVC | Size: 1360 | Accuracy: 0.9729 SVC | Size: 1410 | Accuracy: 0.9729 SVC | Size: 1460 | Accuracy: 0.9729 SVC | Size: 1510 | Accuracy: 0.9729 SVC | Size: 1560 | Accuracy: 0.9714 SVC | Size: 1610 | Accuracy: 0.9729 SVC | Size: 1660 | Accuracy: 0.9729 SVC | Size: 1710 | Accuracy: 0.9700 SVC | Size: 1760 | Accuracy: 0.9700 SVC | Size: 1810 | Accuracy: 0.9700 SVC | Size: 1860 | Accuracy: 0.9714 SVC | Size: 1910 | Accuracy: 0.9714 SVC | Size: 1960 | Accuracy: 0.9700 SVC | Size: 2010 | Accuracy: 0.9714 SVC | Size: 2060 | Accuracy: 0.9714 SVC | Size: 2110 | Accuracy: 0.9700 SVC | Size: 2160 | Accuracy: 0.9700 SVC | Size: 2210 | Accuracy: 0.9700 SVC | Size: 2260 | Accuracy: 0.9700 SVC | Size: 2310 | Accuracy: 0.9700 SVC | Size: 2360 | Accuracy: 0.9700 SVC | Size: 2410 | Accuracy: 0.9700 SVC | Size: 2460 | Accuracy: 0.9700 SVC | Size: 2510 | Accuracy: 0.9700 SVC | Size: 2560 | Accuracy: 0.9700 SVC | Size: 2610 | Accuracy: 0.9700 SVC | Size: 2660 | Accuracy: 0.9700 SVC | Size: 2710 | Accuracy: 0.9700 SVC | Size: 2760 | Accuracy: 0.9700 DecisionTree | Size: 10 | Accuracy: 0.5371 DecisionTree | Size: 60 | Accuracy: 0.7886 DecisionTree | Size: 110 | Accuracy: 0.7643 DecisionTree | Size: 160 | Accuracy: 0.8114 DecisionTree | Size: 210 | Accuracy: 0.8414 DecisionTree | Size: 260 | Accuracy: 0.8529 DecisionTree | Size: 310 | Accuracy: 0.8657 DecisionTree | Size: 360 | Accuracy: 0.8900 DecisionTree | Size: 410 | Accuracy: 0.8514 DecisionTree | Size: 460 | Accuracy: 0.8714 DecisionTree | Size: 510 | Accuracy: 0.8557 DecisionTree | Size: 560 | Accuracy: 0.8971 DecisionTree | Size: 610 | Accuracy: 0.8986 DecisionTree | Size: 660 | Accuracy: 0.9029 DecisionTree | Size: 710 | Accuracy: 0.9000 DecisionTree | Size: 760 | Accuracy: 0.9129 DecisionTree | Size: 810 | Accuracy: 0.9057 DecisionTree | Size: 860 | Accuracy: 0.9114 DecisionTree | Size: 910 | Accuracy: 0.9129 DecisionTree | Size: 960 | Accuracy: 0.9271 DecisionTree | Size: 1010 | Accuracy: 0.8971 DecisionTree | Size: 1060 | Accuracy: 0.9114 DecisionTree | Size: 1110 | Accuracy: 0.9057 DecisionTree | Size: 1160 | Accuracy: 0.9129 DecisionTree | Size: 1210 | Accuracy: 0.9100 DecisionTree | Size: 1260 | Accuracy: 0.9143 DecisionTree | Size: 1310 | Accuracy: 0.9000 DecisionTree | Size: 1360 | Accuracy: 0.9143 DecisionTree | Size: 1410 | Accuracy: 0.9086 DecisionTree | Size: 1460 | Accuracy: 0.9129 DecisionTree | Size: 1510 | Accuracy: 0.9129 DecisionTree | Size: 1560 | Accuracy: 0.9100 DecisionTree | Size: 1610 | Accuracy: 0.9200 DecisionTree | Size: 1660 | Accuracy: 0.9086 DecisionTree | Size: 1710 | Accuracy: 0.9129 DecisionTree | Size: 1760 | Accuracy: 0.9171 DecisionTree | Size: 1810 | Accuracy: 0.9129 DecisionTree | Size: 1860 | Accuracy: 0.9143 DecisionTree | Size: 1910 | Accuracy: 0.9286 DecisionTree | Size: 1960 | Accuracy: 0.9314 DecisionTree | Size: 2010 | Accuracy: 0.9229 DecisionTree | Size: 2060 | Accuracy: 0.9300 DecisionTree | Size: 2110 | Accuracy: 0.9229 DecisionTree | Size: 2160 | Accuracy: 0.9229 DecisionTree | Size: 2210 | Accuracy: 0.9200 DecisionTree | Size: 2260 | Accuracy: 0.9200 DecisionTree | Size: 2310 | Accuracy: 0.9214 DecisionTree | Size: 2360 | Accuracy: 0.9243 DecisionTree | Size: 2410 | Accuracy: 0.9229 DecisionTree | Size: 2460 | Accuracy: 0.9186 DecisionTree | Size: 2510 | Accuracy: 0.9271 DecisionTree | Size: 2560 | Accuracy: 0.9243 DecisionTree | Size: 2610 | Accuracy: 0.9243 DecisionTree | Size: 2660 | Accuracy: 0.9214 DecisionTree | Size: 2710 | Accuracy: 0.9257 DecisionTree | Size: 2760 | Accuracy: 0.9200 RandomForest | Size: 10 | Accuracy: 0.7586 RandomForest | Size: 60 | Accuracy: 0.8743 RandomForest | Size: 110 | Accuracy: 0.8671 RandomForest | Size: 160 | Accuracy: 0.8986 RandomForest | Size: 210 | Accuracy: 0.9100 RandomForest | Size: 260 | Accuracy: 0.9200 RandomForest | Size: 310 | Accuracy: 0.9314 RandomForest | Size: 360 | Accuracy: 0.9443 RandomForest | Size: 410 | Accuracy: 0.9386 RandomForest | Size: 460 | Accuracy: 0.9471 RandomForest | Size: 510 | Accuracy: 0.9457 RandomForest | Size: 560 | Accuracy: 0.9471 RandomForest | Size: 610 | Accuracy: 0.9586 RandomForest | Size: 660 | Accuracy: 0.9514 RandomForest | Size: 710 | Accuracy: 0.9486 RandomForest | Size: 760 | Accuracy: 0.9571 RandomForest | Size: 810 | Accuracy: 0.9529 RandomForest | Size: 860 | Accuracy: 0.9543 RandomForest | Size: 910 | Accuracy: 0.9614 RandomForest | Size: 960 | Accuracy: 0.9557 RandomForest | Size: 1010 | Accuracy: 0.9557 RandomForest | Size: 1060 | Accuracy: 0.9557 RandomForest | Size: 1110 | Accuracy: 0.9543 RandomForest | Size: 1160 | Accuracy: 0.9586 RandomForest | Size: 1210 | Accuracy: 0.9557 RandomForest | Size: 1260 | Accuracy: 0.9557 RandomForest | Size: 1310 | Accuracy: 0.9543 RandomForest | Size: 1360 | Accuracy: 0.9557 RandomForest | Size: 1410 | Accuracy: 0.9543 RandomForest | Size: 1460 | Accuracy: 0.9529 RandomForest | Size: 1510 | Accuracy: 0.9571 RandomForest | Size: 1560 | Accuracy: 0.9529 RandomForest | Size: 1610 | Accuracy: 0.9571 RandomForest | Size: 1660 | Accuracy: 0.9543 RandomForest | Size: 1710 | Accuracy: 0.9529 RandomForest | Size: 1760 | Accuracy: 0.9571 RandomForest | Size: 1810 | Accuracy: 0.9571 RandomForest | Size: 1860 | Accuracy: 0.9600 RandomForest | Size: 1910 | Accuracy: 0.9557 RandomForest | Size: 1960 | Accuracy: 0.9543 RandomForest | Size: 2010 | Accuracy: 0.9586 RandomForest | Size: 2060 | Accuracy: 0.9543 RandomForest | Size: 2110 | Accuracy: 0.9557 RandomForest | Size: 2160 | Accuracy: 0.9586 RandomForest | Size: 2210 | Accuracy: 0.9571 RandomForest | Size: 2260 | Accuracy: 0.9571 RandomForest | Size: 2310 | Accuracy: 0.9557 RandomForest | Size: 2360 | Accuracy: 0.9586 RandomForest | Size: 2410 | Accuracy: 0.9557 RandomForest | Size: 2460 | Accuracy: 0.9557 RandomForest | Size: 2510 | Accuracy: 0.9586 RandomForest | Size: 2560 | Accuracy: 0.9600 RandomForest | Size: 2610 | Accuracy: 0.9600 RandomForest | Size: 2660 | Accuracy: 0.9571 RandomForest | Size: 2710 | Accuracy: 0.9586 RandomForest | Size: 2760 | Accuracy: 0.9600 KNN | Size: 10 | Accuracy: 0.6429 KNN | Size: 60 | Accuracy: 0.9243 KNN | Size: 110 | Accuracy: 0.9386 KNN | Size: 160 | Accuracy: 0.9486 KNN | Size: 210 | Accuracy: 0.9514 KNN | Size: 260 | Accuracy: 0.9557 KNN | Size: 310 | Accuracy: 0.9557 KNN | Size: 360 | Accuracy: 0.9514 KNN | Size: 410 | Accuracy: 0.9571 KNN | Size: 460 | Accuracy: 0.9586 KNN | Size: 510 | Accuracy: 0.9571 KNN | Size: 560 | Accuracy: 0.9586 KNN | Size: 610 | Accuracy: 0.9571 KNN | Size: 660 | Accuracy: 0.9600 KNN | Size: 710 | Accuracy: 0.9600 KNN | Size: 760 | Accuracy: 0.9614 KNN | Size: 810 | Accuracy: 0.9586 KNN | Size: 860 | Accuracy: 0.9586 KNN | Size: 910 | Accuracy: 0.9571 KNN | Size: 960 | Accuracy: 0.9571 KNN | Size: 1010 | Accuracy: 0.9571 KNN | Size: 1060 | Accuracy: 0.9586 KNN | Size: 1110 | Accuracy: 0.9614 KNN | Size: 1160 | Accuracy: 0.9629 KNN | Size: 1210 | Accuracy: 0.9643 KNN | Size: 1260 | Accuracy: 0.9629 KNN | Size: 1310 | Accuracy: 0.9629 KNN | Size: 1360 | Accuracy: 0.9586 KNN | Size: 1410 | Accuracy: 0.9600 KNN | Size: 1460 | Accuracy: 0.9614 KNN | Size: 1510 | Accuracy: 0.9614 KNN | Size: 1560 | Accuracy: 0.9614 KNN | Size: 1610 | Accuracy: 0.9643 KNN | Size: 1660 | Accuracy: 0.9629 KNN | Size: 1710 | Accuracy: 0.9614 KNN | Size: 1760 | Accuracy: 0.9586 KNN | Size: 1810 | Accuracy: 0.9586 KNN | Size: 1860 | Accuracy: 0.9600 KNN | Size: 1910 | Accuracy: 0.9586 KNN | Size: 1960 | Accuracy: 0.9586 KNN | Size: 2010 | Accuracy: 0.9586 KNN | Size: 2060 | Accuracy: 0.9586 KNN | Size: 2110 | Accuracy: 0.9586 KNN | Size: 2160 | Accuracy: 0.9586 KNN | Size: 2210 | Accuracy: 0.9586 KNN | Size: 2260 | Accuracy: 0.9586 KNN | Size: 2310 | Accuracy: 0.9586 KNN | Size: 2360 | Accuracy: 0.9586 KNN | Size: 2410 | Accuracy: 0.9586 KNN | Size: 2460 | Accuracy: 0.9586 KNN | Size: 2510 | Accuracy: 0.9586 KNN | Size: 2560 | Accuracy: 0.9586 KNN | Size: 2610 | Accuracy: 0.9586 KNN | Size: 2660 | Accuracy: 0.9586 KNN | Size: 2710 | Accuracy: 0.9586 KNN | Size: 2760 | Accuracy: 0.9586
Final evaluation¶
In [15]:
# Get the model with best accuracy by choosing the maximum index 1 in the nested list
best_model = max(models, key = lambda x: x[1])
print(f"{best_model[0]} | Size: {best_model[2]} | Accuracy: {best_model[1]}")
SVC(C=11) | Size: 1360 | Accuracy: 0.9728571428571429
This is our best Model:
SVC with C = 11
Size on dataset 1360
With an Accuracy 97.29%,
The accurcy will get higher on the kaggle submission.
Kaggle submission¶
In [16]:
# Our best model.
model = best_model[0]
# Getting our test data and dropping useless features.
df_test = pd.read_csv("./assets/test.csv", index_col = 0)
df_test = df_test.drop(columns = useless_features)
# Scaling
X_test2_scaled = scaler.transform(df_test)
# Taken from the kaggle submission side.
y_test2 = model.predict(X_test2_scaled)
y_test2 = pd.DataFrame(y_test2, columns=["Quality"])
y_test2.index.name = "ID"
y_test2[['Quality']].to_csv("submission.csv")